Five writing prompts you can copy straight into ChatGPT, Claude or Gemini, each with the reasoning behind it so you can adapt it to your own voice and subject.
5 copy-ready prompts
Works in any chat model
Editing and drafting
The prompts
Prompt 1
Draft in my voice, not the model default
First drafts of posts, newsletters and updates that should sound like you.
You are helping me draft a [type of piece] for [audience].
Here are three samples of my own writing:
"""
[paste 200-400 words you have written]
"""
Study the sentence length, vocabulary and level of formality in those samples. Then write a [word count] draft on: [topic].
Rules:
- Match my rhythm and vocabulary, not a generic professional tone.
- No introductory throat-clearing. Start with the substance.
- No phrases like "in today's fast-paced world" or "unlock the power of".
- Where you are unsure of a fact, mark it [CHECK] instead of inventing detail.
Why it works: The voice samples do most of the work. Without them, every model drifts to the same flat register.
Prompt 2
Edit for clarity without flattening the writing
Tightening a draft you already like.
Edit the text below for clarity and flow. Keep my voice: do not make it more formal, and do not replace concrete words with abstract ones.
For each change, list it as: original phrase -> your version -> one-line reason.
Then give me the full edited text.
Text:
"""
[paste your draft]
"""
Why it works: Asking for the change list first stops the model quietly rewriting everything and lets you accept edits selectively.
Prompt 3
Outline from a messy brain dump
When you know what you want to say but not in what order.
Below is an unstructured set of notes. Turn it into a logical outline for a [format] aimed at [audience].
For each section give me: a heading, the single point it makes, and which of my notes it uses.
At the end, list anything important the notes are missing, and anything I have said twice.
Notes:
"""
[paste your notes]
"""
Why it works: The "what is missing" question is the valuable part - it catches gaps you cannot see because you already know the subject.
Prompt 4
Rewrite for a specific reader
Adapting one piece for a different audience or channel.
Rewrite the text below for this reader: [describe them - role, what they already know, what they care about, how much time they have].
Constraints:
- Keep every factual claim unchanged.
- Cut anything this reader would consider obvious.
- Lead with what matters to them, not with background.
- Target length: [number] words.
Text:
"""
[paste your text]
"""
Why it works: Describe the reader by what they already know. "Non-technical" is vague; "a finance manager who has never used an API" is actionable.
Prompt 5
Interrogate my argument before I publish
Opinion pieces, proposals and anything you will be challenged on.
Read the argument below and respond as a smart, sceptical reader who disagrees with me.
Give me:
1. The three strongest objections, in the order a critic would raise them.
2. Any claim I have made that is not actually supported by the evidence I gave.
3. The weakest sentence in the piece and why it is weak.
Do not soften your answers or praise the writing.
Argument:
"""
[paste your piece]
"""
Why it works: Run this before you publish rather than after. It is uncomfortable and far cheaper than being corrected in public.
How to get writing you would actually publish
The gap between a mediocre AI draft and a useful one is almost never the model. It is the amount of specific material you put into the prompt before asking for anything.
Why generic prompts produce generic writing
A language model answers with the most statistically unremarkable version of what you asked for. Ask for "a blog post about productivity" and you get the average of every productivity post ever written, because that is literally what the request describes. The prompts on this page work by narrowing that space: they supply your voice samples, your audience, your constraints and your actual material, so the average the model reaches for is a much smaller and more specific one.
The four things worth specifying every time
Audience - not "general readers" but what they know, what they care about, and how long they will give you.
Voice - the fastest route is pasting a few hundred words you have already written rather than describing your style in adjectives.
Length and format - a word count and a structure prevent the padded five-paragraph shape that every model defaults to.
Prohibitions - naming the clichés you never want to see is more effective than asking for "engaging" writing.
Editing prompts beat drafting prompts
Most people reach for AI at the blank page, which is the moment it adds least. A model has no idea what you want to say, so it invents something plausible and you spend longer rewriting it than you would have spent writing. The higher-leverage use is the opposite end: you write the messy version, then use the model to tighten, restructure and challenge it. The material is yours, so the output keeps your thinking and gains a second pair of eyes.
Iterating instead of re-prompting
When a draft comes back wrong, most people rewrite the whole prompt and start again. It is usually faster to respond with a correction: "the second section is too abstract - rewrite it using the example from my notes" or "this sounds like marketing copy, make it plainer". The conversation retains everything you established earlier, so each correction is cheap. Starting over throws that away.
It also helps to ask for options rather than a final answer. "Give me three different openings, each taking a different angle" gives you something to choose between, which is far easier than judging a single draft in isolation.
Fact-checking is not optional
Models produce confident, fluent, wrong statements - names, dates, statistics and citations that look exactly like real ones. The instruction to mark uncertain claims with [CHECK] in the first prompt is a small guardrail, not a guarantee. Anything factual that appears in your final draft needs a source you have actually looked at. This matters twice over if you publish: search engines and readers both treat invented facts harshly, and a single fabricated statistic undermines everything around it.
Keeping your own voice intact
The real risk of writing with AI is not being caught - it is sounding like everyone else who uses the same tool. Protect the parts that are yours: the specific example only you have, the opinion you are willing to defend, the sentence that sounds like speech. Use the model for structure, tightening and pressure-testing, and keep authorship of the ideas.
Related prompt packs
For campaign copy and positioning, see the AI Marketing Prompts. Developers writing documentation as well as code will want the AI Coding Prompts. If the writing in question is a job application, the AI Resume Prompts cover that specific format, and the AI Study Prompts help when you are writing to learn rather than to publish.
Frequently asked questions
Do these prompts work in Claude and Gemini as well as ChatGPT?
Yes. They rely on clear instructions and supplied context rather than model-specific tricks, so they work in any modern chat assistant.
How do I stop AI writing from sounding generic?
Paste samples of your own writing, name the phrases you never want used, specify the audience precisely, and use the model to edit your draft rather than to produce one from nothing.
